Using the 3D model RAPID to invert the shoot dieback ratio of vertically heterogeneous Yunnan pine forests to detect beetle damage

Abstract

As a significant forest disturbance factor, pine shoot beetles (PSB) have caused widespread tree mortality in Yunnan pine forests over southwestern China. Due to significant vertical heterogeneity of tree crowns, empirical remote sensing methods or one-dimensional (1D) radiative transfer (RT) models are not qualified enough to accurately retrieve the damage level of PSB damage. Therefore, three-dimensional (3D) RT models which have the ability to describe vertically complex heterogeneous canopies are more suitable for the damaged forests. In this study, shoot dieback ratio (SDR) was used to parameterize the Radiosity Applicable to Porous IndiviDual Objects (RAPID) model, in order to accurately estimate the PSB damage. Specifically, the vertically heterogeneous characteristics of the biophysical and optical properties within the canopy were carefully considered. First, we developed an approach to reconstruct damaged forest scenes through individual tree segmentation from UAV-lidar data and in situ measurements. Second, the simulated reflectance of forest scenes was compared with the actual hyperspectral and Sentinel-2 (S2) reflectance data to evaluate the performance of the parameterized RAPID. Then, the parameterized RAPID model was used to create a look-up table (LUT) containing the leaf chlorophyll content (Cab), effective leaf area index (LAIe), SDR, and dead needles LAIe (LAIdead, LAIe × SDR) as well as the corresponding simulated canopy reflectance. Furthermore, tree allometric equations were used to regularize the LUT with the intention to improve the estimation accuracy. Good agreement (RMSE: 0.015-0.031, index of agreement: 0.97-0.99) was found between the simulated reflectance and the remote sensing observations. The retrieval accuracy of the biochemical and biophysical parameters using S2 data yielded R2 = 0.55 (normalized RMSE (NRMSE) = 19%) for Cab, R2 = 0.70 (NRMSE = 15%) for the LAIe, R2 = 0.66 (NRMSE = 25%) for SDR, and R2 = 0.82 (NRMSE = 18%) for LAIdead. Using hyperspectral data with higher spatial and spectral resolution yielded better retrieval performance. This study demonstrated that 3D models (such as RAPID) have great potential to accurately invert SDR and LAIdead to assess PSB damage for the forests.